Skill manifests and required skills can influence agent behavior and may reference executable resources. Treat untrusted skills as untrusted code and instructions.
Report suspected vulnerabilities privately through GitHub's security advisory workflow for this repository.
Version 0.3.x receives security fixes.
Consumers should:
- execute only immutable commits or verified content digests;
- verify tags against declared commits;
- review targets, extensions, and package requirements;
- retain a previous known-good pin;
- never place credentials in
skill.json.